基于核函数的有监督哈希视频图像检索

Supervised Hashing Video Image Retrieval with Kernels

在公安的电子物证检验工作中,经常会获取大量的影像、图像等信息,需要将这些信息与已掌握的视频图像内容进行比对,发现有价值的线索、证据。如果仅依靠人工检索,工作量巨大,耗时耗力,效率极低。这就需要开发出一套智能检索系统,通过使用一种高质量和高效率的大规模视频图像检索方法,准确、有效的关联出所需的视频图像内容。目前,常用的视频图像检索方法是哈希方法,被用来做相似性计算和检索。然而,现存的检索方法或者是缺乏足够好的性能,或者是陷入复杂的模型学习之中。文章提出了一种新的基于核函数的哈希模型应用于视频图像的人脸检索,该模型仅仅需要少量的监督信息,却能够在可行的训练时间下,获得高质量的哈希,有效提高检索质量和效率。

Inspection of electronic evidence in public security, we often get a lot of videos images and other informa- tion.We need to compare this mastered information content of the videos and images to find valuable clues and eviden- ce. If you only rely on manual searches, it will be a timeconsuming, inefficient work.Thus we requires an intelligent retrieval system, through the use of a high quality and high efficiency of large scale video image retrieval method, At present , the commonly used video retrieval method is hash , However, the existing search methods or lack of adequate performance, or being caught in a complex learning model. So the model of supervised hashing with kernels h as been used in this paper. The model requires only a small amount of supervision information, but can obtain high- quality hash in practical training time which can improve the retrieval quality and efficiency.